Spike score share gave no clear gain in seizure onset zone ranking for 18 SEEG sleep records with STE events and a fixed model

We tested whether released spike score share adds seizure onset zone (SOZ) rank information in public stereoelectroencephalography (SEEG) data. The test used the first hour of 18 sleep records from OpenNeuro ds005398 version 1.1.3. The baseline model included event count, contact number, event duration, and mean artifact score. The expanded model added a smoothed share of events with spike score above 0.5. Each test patient was excluded from both new model fits. The main short-time energy (STE) test used 1,285 channels and 18,801 events. We measured rank quality with area under the receiver operating characteristic curve (AUC). Mean AUC changed from 0.7445 to 0.7354. The gain was -0.0091, with a descriptive 95% full-refit patient bootstrap interval of [-0.0361, 0.0144]. Four fixed STE checks also gave no clear rank gain. The fixed MNI detector check gave a gain of 0.0322, with interval [0.0035, 0.0993]. This was a secondary test in the same cohort. Upstream score model training overlap is unknown. The results do not establish external or clinical validity. The package includes source tables, pinned retrieval rules, code, patient predictions, all 7,000 bootstrap draws, independent checks, and an exact full numerical repeat.

high-frequency oscillations · seizure onset zone · stereoelectroencephalography · spike score · event detection · patient bootstrap